In order to solve the security problems caused by network vulnerabilities, a web application vulnerability detection method based on machine learning is proposed to effectively prevent cross site scripting attacks of web applications and reduce the occurrence of network security incidents. Through the in-depth study of the existing security vulnerability detection technology, combined with the development process of machine learning security vulnerability detection technology, the requirements of security vulnerability detection model are analyzed in detail, and a cross site scripting security vulnerability detection model for web application is designed and implemented. Based on the existing network vulnerability detection technology and tools, the verification code identification function is added, which solves the problem that the data can be submitted to the server only by inputting the verification code. According to the server filtering rules, the network code bypassing the server filtering is constructed. Experimental results show that the model has a low rate of missed detection and false alarm, and the improved model is more efficient.